Set the caller's pointer to NULL in the destroy functions

[karpovantonme]

Closes #2387

t8_stash_destroy and t8_cmesh_trees_destroy both take a pointer to the caller's handle so that they can clear it after freeing, and both headers say so:

/** Free all memory associated in a stash structure.
 * \param [in,out]  pstash  A pointer to the stash to be destroyed.
 *                  The pointer is set to NULL after the function call.
 */

/** \param [in,out]  trees The tree structure to be destroyed. Set to NULL on output. */

Neither does it. Both assign to the parameter instead of dereferencing it:

T8_FREE (stash);
pstash = NULL;      /* clears the local copy, not the caller's handle */

T8_FREE (trees);
ptrees = nullptr;

So the caller is left holding a dangling pointer, and the whole reason for taking a double pointer is lost.

Why this is not only cosmetic

The callers are written as if the contract held.

t8_cmesh_commit.cxx:626 guards the destroy on the handle being non-NULL:

if (cmesh->stash != nullptr) {
  t8_stash_destroy (&cmesh->stash);
}

and line 145 of the same file reads through the same guard:

if (cmesh->stash != nullptr && cmesh->stash->classes.elem_count > 0) {

Since cmesh->stash is never set to NULL, that second check passes on a freed pointer and dereferences it. t8_cmesh_helpers.cxx:263-264 and t8_cmesh_readmshfile.cxx:1596 read cmesh->stash->... as well.

t8_cmesh_bcast (t8_cmesh.cxx:792) is the clearest case: it destroys the stash and sets cmesh_out->committed = 1, and the cmesh keeps living with a dangling stash.

In the teardown path (t8_cmesh_destroy) the dangling pointer is harmless, since the cmesh is freed right after. It is the commit and bcast paths that matter.

The change

Two characters, one in each function: pstash = NULL becomes *pstash = NULL, ptrees = nullptr becomes *ptrees = nullptr. Nothing else touched.

I have not added a regression test. The natural one would assert that cmesh->stash == NULL after commit, but I did not want to guess where it belongs in your test layout. Happy to add it if you point me at the right file.

Found with a cppcheck --enable=warning sweep over src/ (uselessAssignmentPtrArg at t8_cmesh_stash.c:70 and t8_cmesh_trees.cxx:1295), then traced through the callers by hand.

Author file added as doc/author_karpov.txt per CONTRIBUTING, and both commits are signed off.